← Back to projects

HELM

AI Orchestration with Process Engineering

Most agent frameworks are ad-hoc function chains. They work for demos but break in production — no error recovery, no audit trail, no governance. When Agent B fails, Agent A doesn't know. No one knows what happened or why.

The Approach

HELM brings three decades of business process engineering to agentic AI. Instead of inventing new orchestration patterns, it applies proven standards — BPMN for structured workflows, CMMN for adaptive case management, and DMN for decision governance. One engine, configured per scenario.

Architecture Explorer

BPMN

Workflows

Deterministic process orchestration for known sequences

CMMN

Cases

Adaptive case management for dynamic, non-linear work

DMN

Decisions

Auditable decision tables governing agent behavior

Choose a scenario

A multi-phase workflow with parallel planning agents and gateway synchronization — orchestrated by BPMN.

Phase 1 — Research

Phase 2 — Planning (parallel)

Phase 3 — Synthesis & Quality

Phase 4 — Build & Ship

Click any node to explore how HELM orchestrates that step.

Tech Stack

Language

TypeScript

Workflow Engine

BPMN 2.0 Runtime

Case Engine

CMMN 1.1 Runtime

Decision Engine

DMN 1.3 Tables

Agent Integration

LLM APIs (multi-provider)

Architecture

Event-driven, modular

Key Decisions

Why BPMN over ad-hoc chaining?

BPMN gives you error boundaries, retry policies, parallel gateways, and compensation handlers out of the box. Building these from scratch in every agent framework is duplicated effort with worse results.

Why CMMN for edge cases?

Real-world tasks aren't always sequential. CMMN models work that adapts — discretionary tasks, condition-based activation, and milestone-driven completion. The case handles uncertainty; the workflow handles certainty.

Why DMN for governance?

Decision tables are auditable, versioned, and readable by non-engineers. When you need to change which model handles a task or when human approval is required, you update a table — not code.